Research On Improvement Of Particle Swarm Algorithm Based On Momentum

Particle swarm optimization(PSO)is a population intelligent algorithm proposed by kenndey et al.In 1995 based on the observation of fish and bird behavior.Its idea comes from evolutionary computation theory.By simulating the foraging behavior of birds,the birds can reach the optimal foraging point through group cooperation.Particle swarm optimization algorithm has the advantages of fast convergence and simple structure,so it has been recognized,discussed and improved by many scholars,and plays an important role in more and more fields.However,due to its own characteristics of random search,there are some problems such as premature convergence and slow iteration speed in the later stage.This paper compares particle swarm optimization(PSO)and classical momentum algorithm,which belong to the same branch of optimization algorithm,and finds that they have high similarity.It can be used for reference to the improvement of classical momentum algorithm in PSO optimization.Therefore,this paper proposes the improvement of particle swarm optimization algorithm based on momentum to solve the problems of traditional particle swarm optimization algorithm,such as easy to fall into local optimization and slow convergence speed.(1)Traditional particle swarm optimization algorithm uses the same inertia weight for all particles and ignores the characteristics of a single particle,resulting in low convergence accuracy and easy to fall into local optimum.Combined with the strategy of setting each dimension adaptively in RMSprop algorithm,an adaptive inertia weight particle swarm optimization algorithm RMSPSO is proposed.Considering the velocity and momentum of each dimension of particles,the algorithm sets the adaptive dynamic inertia weight,which makes the algorithm achieve a good balance between global optimization and local optimization.Ten typical test functions are selected,and the improved particle swarm optimization algorithm(RMSPSO)is compared with four mainstream particle swarm optimization algorithms.The results show that the RMSPSO algorithm has made great progress in convergence speed and accuracy in unimodal,multimodal and combined functions.(2)Traditional particle swarm optimization(PSO)tends to oscillate near the best point in the middle and later iterations,which leads to the problem of slow convergence speed.The convergence process of traditional particle swarm optimization(PSO)algorithm is analyzed.It is found that the current stride of particle update is too large due to the velocity accumulation term in the process of particle velocity update,which leads to the phenomenon that the particle position crosses the local best point and vibrates.In this paper,a particle swarm optimization algorithm based on nag is proposed,which is based on the idea of Nag algorithm to improve the classical momentum algorithm.Use the position of particle "prospect" to influence the current update,so that the particle will decelerate before reaching the optimal value to avoid oscillation.Ten typical test functions are selected,and the improved particle swarm optimization(NAGPSO)algorithm is compared with three mainstream particle swarm optimization algorithms.The results show that NAGPSO improves the optimization ability of the algorithm to a certain extent,especially in the multimodal problem.(3)Using the idea of adaptive learning rate setting strategy Adam algorithm,the concept of momentum in physics is introduced into the strategy of particle swarm optimization adaptive setting inertia weight.An adaptive particle swarm optimization algorithm for inertia weight based on Adam is proposed.The current gradient information of particles is updated according to momentum,and the gradient square value is calculated according to the exponential weighted average.The inertia weight is set adaptively according to the offset corrected value of gradient information and gradient square value.Ten typical test functions are selected,and the improved particle swarm optimization algorithm(ADAMPSO)is compared with four mainstream particle swarm optimization algorithms.The results show that ADAMPSO improves the convergence speed,accuracy and stability of the algorithm.